A tool to measure complexity in public health interventions: Its statistical properties and meta-regression approach to adjust it in meta-analysis

BackgroundPublic health interventions are conventionally cited as a popular example of complex interventions. Complexity of public health interventions has always been an obstacle for the meta-analysis of these studies. Earlier an attempt was made by Public Health Evidence South Asia (PHESA) to numerically measure the complexity in public health interventions by a tool. This study is an extension of that initiative which aims to investigate various statistical distributional properties of the complexity score and adjust the measured complexity in meta-analysis.MethodsComplexity score of 71 studies was used to identify the best probability distribution that fits the complexity score, study its sampling distribution and determine the optimum power transformation. Meta-regression was employed to adjust the measured complexity in meta-analysis.ResultsLognormal distribution was observed to be an ideal probability distribution for the complexity score, the sampling distribution of the mean of complexity score was found to be normally distributed and the optimum power transformation for the complexity score was ‘−0.42’. The raw estimate from random effects meta-analysis was found to be −0.05, 95% CI −0.14 to 0.04, whereas the estimate adjusted for complexity from meta-regression was −0.048, 95% CI −0.13 to 0.03. There was a reduction in the proportion of heterogeneity (I – squared) after adjusting for complexity (73.01%–66.30%), indicating that complexity had an impact on the effect estimates of studies.ConclusionThe concept of measuring the inherent complexity and adjusting it in meta-analysis adds novelty to the existing meta-analysis approach. This innovative approach is likely to create a new dimension for meta-analysis of complex community level interventions and provide more precise evidence. However, further methodological research and piloting is required to establish the validity and sensitivity of this approach.